Our interview with Andrea Volpini of WordLift reveals the strategies, metrics and editorial choices that allow a company to enter the answer flows generated by ChatGPT, Claude, Gemini (and soon Google AI Mode too)
SEO is no longer an optimization exercise aimed at showing up in the SERPs: with the arrival of AI Mode, query fan-out and language models, the challenge has shifted to new ground, where visibility depends on how content is understood, broken apart and reassembled by machines.
It is a transformation that forces us to rethink the very concept of “ranking” and to come to terms with a web made of semantic blocks, citations and automated reasoning.
In this scenario, writing well-structured articles or detailed product pages is no longer enough: you need to build a body of knowledge that AI can read, capable of fitting into its logical pathways. It is a shift that may scare many entrepreneurs and marketers, but it opens up enormous possibilities for those who can interpret the change with vision and method.
For this new installment of SEO Confidential we met Andrea Volpini, CEO and co-founder of WordLift. An innovator and entrepreneur with twenty years of international experience in marketing and digital publishing, Volpini has dedicated his career to the semantic web and artificial intelligence.
After founding companies such as InSideOut10 and InsideOut Today in Egypt, he now leads WordLift, one of the most advanced players in building knowledge graphs and solutions to make content “machine-visible”.
With him we discussed the hottest topics of the moment: from the concept of the Reasoning Web – a web that no longer merely displays pages but builds actual logical pathways – to the creation of LLMifiable content, designed to be understood and reused by artificial intelligence.
We talked about how to measure a brand’s machine-visibility, the new metrics that really matter in AI Mode, and the strategies that allow an SMB to become the source cited and recognized by systems like ChatGPT, Gemini or Claude.
What came out is an interview that breaks the mold: concrete, visionary and surprising. A conversation that shows why the SEO of the future will no longer be a race to climb the SERPs, but a battle to become the indispensable piece in the machines’ reasoning.
Machine-visibility, authority and “LLMifiable” content: Andrea Volpini on the new era of SEO in AI Mode
Many entrepreneurs fear that AI Mode will reduce direct traffic to websites, because users already find the answer inside Google: how can an SMB turn this scenario from a risk into an opportunity, and which metrics should it monitor to understand whether it is truly benefiting from this new form of search?
For an SMB the point is not “losing clicks”, but gaining relevance in the AI context. The opportunity is to become the authoritative source that a search engine or a language model cites, uses, recommends. I imagine a web without static pages, made exclusively of answers provided by AI and presented through a dynamic interface that adapts to the user.
There are no clicks. There are concepts, associations, strong brands, answers and solutions that derive from constantly evolving semantic associations and heuristics (what is the user feedback? what information can I gather about this product compared to a competing product? etc.).
To achieve this authority you need a core of content that is first of all LLMifiable: machine-readable (schema.org, knowledge graphs, text chunking, synthetic examples, questions and answers in open datasets, llms.txt and more) and content that answers users’ real questions in depth.
A few days ago we closed an $18,000 contract with an American client. He found us using ChatGPT. He wants to be as visible to his customers as we were to him. That is the first thing we want to measure. The impact on our business. Then we can define proxy metrics: impressions and citations in AI Mode, assisted clicks (i.e. searches where the user first interacts with the AI and then reaches the site), the salience of our entities in the Google Knowledge Graph, and the crawl logs of AI agents. These indicators show whether we are becoming machine-visible, that is, present and competitive in the new context of AI search.
AI Mode no longer picks an entire page, but specific passages that answer “sub-questions” generated by the engine: what practical strategies should companies adopt to turn their content into blocks the AI can actually use, so they don’t risk being excluded from the answers?
The strategy is simple. Close your eyes: your website has disappeared. What remains is a set of extremely representative semantic blocks, no longer a continuous narrative flow locked inside a static page. Each block must be:
- Self-contained – able to answer a precise question without requiring the reader to go through the entire article.
- Semantically labeled – with markup, concrete concepts, Q&A, tables, lists: elements a model can easily extract and rephrase.
- Connected to the Knowledge Graph – that is, part of a network of entities and relationships that provides context and authority.
Imagine building the Graph RAG (that is, a system for retrieving content from a KG) on behalf of third parties.
An SMB can start simply: turn its FAQs into machine-readable datasets, break long texts into paragraphs optimized as “chunks”, add practical examples and comparisons with competing products. Link these blocks together, avoid duplication, sharpen the targeting.
In this scenario the winner is not whoever writes more articles, but whoever offers the best knowledge modules to feed the AI’s dynamic answers. This reduces the risk of being excluded and increases the likelihood of being cited as an authoritative source.
Today, as we know, “focusing on keywords” is no longer enough; you need to help the AI in its reasoning: how does that translate into concrete editorial choices for guides, product pages or industry articles?
Keywords used to help us get found by an engine that matched queries and documents. Today content has to speak to a system that reasons in steps: AI Mode breaks questions down into sub-questions, connects entities and builds logical pathways.
I call this new phase the Reasoning Web. Years ago, in the pre-ChatGPT era, I urged marketers to become prompt masters, experimenting with the first open source models like Google’s T5. Today the next step is to think in terms of the models’ memory and learning.
A model learns through examples – exactly as we sapiens do. That is why I talk about LLMifiable content (which is a horrid term, borrowed from English and mangled, but it gets the idea across): information accompanied by clear examples, use cases, correct and incorrect comparisons. This is what feeds training policies (from supervised instruction all the way to reinforcement learning). In other words, every piece of content we produce can become educational material for the AI.
The editorial translation is concrete:
- Guides → become decision maps: not just explaining a topic, but anticipating logical pathways, common problems, possible solutions, alternatives compared.
- Product pages → from generic descriptions to answers to why choose this product over a competitor, with structured attributes (material, compatibility, benefits, use cases).
- Industry articles → must offer axioms and evidence: data, statistics, examples, historical comparisons the AI can use to support a line of reasoning.
In practice, every piece of content becomes a learning opportunity. We no longer optimize just for the click, but to train the models themselves: to become a step in the chain of reasoning.
Andrea, the personalization factor seems destined to radically change visibility: two users with the same query can receive very different answers. What kinds of content and semantic signals increase the chances that a brand gets selected as an authoritative source in such personalized scenarios?
Personalization changes everything: two users with the same query receive different answers. If we don’t know who our customer is, we risk being penalized every time.
When, on the other hand, the ICP (Ideal Customer Profile) and personas are clear, personalization becomes a native function of the system. The AI recognizes content designed for that target as the most relevant.
That is why content with a clear target is by definition AI-friendly: concrete examples, FAQs, structured product pages, semantic markup and connections to the Knowledge Graph. These signals help the model understand who that content is meant for and activate it only when needed.
In short: the winner is not whoever speaks to everyone, but whoever builds semantic blocks tailored to their ideal customers.
Query fan-out makes a deterministic approach to ranking impossible: how can a company turn this limitation into a competitive advantage, and what are the most concrete signals to monitor to understand whether its content is actually being intercepted by the sub-queries generated by Google?
With query fan-out Google no longer returns a deterministic list of results, but fragments the question into a series of sub-queries and composes the answer dynamically. This makes it impossible to “rank” as in the past, but it opens up a competitive advantage: we can design granular content that intercepts multiple steps of the reasoning.
In practice, every semantic block on the site must be able to answer a specific sub-question: definitions, comparisons, product attributes, FAQs, examples. The more structured the content is and the more connected to the Knowledge Graph, the higher the probability that the AI will “reuse” it at the right moment.
With WordLift we also put the embeddings of individual blocks into the graph, so we can correlate related content even when it is spread across different pages. Optimization is no longer keyword-centric but reasoning-centric: we design blocks around the logical steps the AI takes, maximizing relevance to the answer pathway. The goal: to become the necessary piece in the AI’s reasoning.
In your posts you often talk about the ontological core as the foundation for staying visible: can you clarify what it is and how a business should build this semantic base in practice, and which resources (among knowledge graphs, schema, product comparisons) are indispensable today to avoid being cut out?
It is the semantic heart that defines a company’s identity in the AI web. It is the set of concepts, entities and relationships that describe who we are, what we offer and why we are relevant.
Without this core, content risks getting scattered: it turns into isolated fragments the AI cannot connect back to the brand. With an ontological core – a term introduced by Tony Seale – every informational module is traced back to a coherent, machine-readable graph.
The construction can start from demand analysis (what problems and needs emerge) or from the existing content the company has already produced. The fundamental exercise is putting yourself in a robot’s shoes: understanding not only which concepts represent us, but above all which relationships connect them and for which audience they are relevant.
The AI Audit evaluates very technical elements such as schema markup, chunking and crawler accessibility: which of these aspects represent the real “quick wins” a company should fix right away in order not to be cut out of the answers generated by AI search?
WordLift’s AI Audit was created precisely to show a business where to intervene right away. Some aspects are more “strategic” and take time; others are quick wins that make an immediate difference.
The three priority interventions today are:
- Correct schema markup and semantic HTML → without markup, AIs don’t know how to interpret products, articles or events. It is the foundation of being readable. Just as it is important to write HTML in a sequential, syntactically correct way.
- Content chunking → breaking overly long texts into semantic blocks, each capable of answering a sub-question. This increases the chances of being reused in AI answers.
- Accessibility for AI crawlers → sitemaps, llms.txt (despite Google’s denials, Ed.) and loading performance: if content is not easily reachable or loads too slowly, the AI simply doesn’t see it.
These are rapid improvements that in many cases can be achieved in a few weeks. The value lies in the fact that every minute your content is not readable by AIs is a minute in which a competitor gains visibility in your place.
An SEO audit used to help you climb the SERPs; an AI audit helps you get among the cited sources: what concrete metrics show whether a site is gaining visibility in the answers of ChatGPT, Gemini or Claude?
An AI Audit helps you understand whether or not your brand makes it into AI-generated answers. The only metric that matters is the impact on the business.
Agentic SEO promises to turn the website into a “living” system, able to adapt in real time to customers’ questions: what are the first concrete steps a company should take to move from a static model to a truly agentic one without risking the loss of editorial control?
In essence, you start from knowledge and proceed with automation.
- Build the Knowledge Graph → map the key entities (brand, products, services, people) and the relationships. It is the memory that allows agents to work consistently on our projects.
- Expose content in machine-readable format → schema.org, semantic chunking, relevance scores. This is the common language that allows AIs to use the site as a source.
- Automate the most repetitive workflows → generating FAQs, product descriptions, markup updates, always with an editorial validation layer. The agent proposes, the company approves. It is a two-horse race: as Ethan Mollick says, we work as centaurs — half human and half machine, each with our own strengths, together faster and smarter.
- Monitor by impact → understand whether content is being intercepted by agents, analyze where assisted clicks land, the ratio between AI crawls and impressions, and above all the concrete impact on revenue.
At this point, I have to ask you: what are, in your view, the most common mistakes that prevent a site from being found and understood by AI-based search engines?
The most common mistakes are always the same ones as in SEO – with the addition, though, of the inability to build semantic maps that are coherent with one another – the absence of semantic signals, the inability to differentiate, poor specialization, generic content without concrete examples.
Again – close your eyes – the website has disappeared. Your customer is in front of an AI. What is truly memorable about what you do? How do you want your brand to be remembered? The most common mistake is not investing in innovation.
Andrea, before saying goodbye, I would like to ask you: what are some practical examples of content or strategies that have already allowed brands or websites to be cited by AI assistants such as ChatGPT, Claude or Perplexity, demonstrating the value of the AI Audit?
Do you have Q&A content without semantic markup? Add it. And reuse that content to train your assistant to answer customers better. One of our clients took years of work on FAQs created with WordLift and “fed” them to the new assistant: today it answers questions precisely and captures the most interesting commercial inquiries.
Remember: when GPT-5 doesn’t have an answer, it searches Bing or Google. And when it is Google answering, our study shows that it is often the content from the People Also Ask block that influences the snippet the model will use. The good news is that we can influence this content directly with FAQ markup applied to relevant, quality content.
Those who master chunking, knowledge graphs and semantic markup will win the AI Search challenge
As clearly emerged from Andrea’s words, a brand’s value is no longer measured in clicks, but in its ability to be recognized as a reference point by artificial intelligence.
Engines don’t show simple results anymore; they reassemble concepts and stories. That is where it is decided whether a business stays present or is progressively pushed into the shadows.
The priority is not to mass-produce content, but to build machine-readable blocks of knowledge that are consistent with the brand’s identity. For SMBs this means going beyond the SERP and staking out their own semantic space: knowledge graphs, LLMifiable content, solid signals that can withstand automated interpretation.
And it is precisely here, in my opinion, that the real challenge lies: SEO is no longer a set of tactics to apply, but a long-term investment in credibility.
Artificial intelligence selects what is clear, structured and coherent, and pushes into the background what is not. Every company must therefore decide whether it wants to be recognized as a reliable source or risk becoming marginal.
SEO thus turns into reputation work: every inconsistency weakens you, every poorly structured piece of content is a missed opportunity. Those who manage to consolidate their semantic identity will be cited, recommended and remembered; those who remain anchored to old logics will inevitably be overtaken by AI-generated summaries.
The change is already underway and it is moving fast.
On this point I have to insist:
you absolutely must be ready if you want to be among the brands the AIs choose to highlight.
It is no longer about chasing rankings, but about building the machine-visibility that brings trust and conversions.
All that is left for me is to warmly thank Andrea Volpini for sharing such complex concepts, so vital for the future of businesses, with precision and depth – along with plenty of practical advice.
SEO Confidential returns next week with another unmissable guest: see you soon!
#fullspeedahead
